Databricks Contributes OpenSharing Protocol to Linux Foundation for Enhanced AI Asset Exchange
On June 10, 2026, Databricks announced its contribution of OpenSharing to the Linux Foundation, marking a significant step in open data sharing for artificial intelligence (AI) assets across various platforms and organizations. OpenSharing is an open protocol that extends upon Delta Sharing, enabling the sharing of models, agent skills, and unstructured data without the need for copies, and introduces support for Apache Iceberg clients. The initiative aims to replace proprietary marketplaces and point-to-point integrations with a common exchange layer for the AI ecosystem.
OpenSharing is positioned as the logical next step in the evolution of Delta Sharing, an existing open standard for data sharing that has been embraced by companies like Amadeus, Atlassian, SAP, Stripe, LSEG, and TheTradeDesk. While Delta Sharing was limited to datasets, OpenSharing expands its reach to encompass AI-specific assets crucial in the age of agents.
This announcement directly addresses a pressing challenge faced by enterprises as they begin to deploy agent architectures: the need for seamless movement of AI assets between systems. Currently, this transfer typically relies on point-to-point integrations or proprietary marketplaces. OpenSharing responds to this need by creating a framework that facilitates interoperability across diverse technical choices in the marketplace, amid a broader reconfiguration of data platforms surrounding AI.
From Data Sharing to AI Asset Sharing
The core mechanism of OpenSharing is built on the principle of “sharing without copying,” inherited from Delta Sharing. This allows a provider to expose their assets, which a recipient can access without physically transferring data, thus avoiding duplication and associated transfer costs. OpenSharing applies this principle to AI assets—models, agent skills, and unstructured datasets—which can be shared under the same governance rules as analytical tables, complete with access controls that determine who can access what and under which conditions.
One of the most significant technical advancements in OpenSharing is its support for Apache Iceberg clients via a REST catalog. The protocol also facilitates direct connections from on-premises and private cloud data sources to cloud platforms, eliminating the need for prior migration. For IT leaders, this compatibility opens up sharing opportunities for recipients who do not utilize the same technology stack, enabling true interoperability between organizations with varying technical choices.
An Open Standard Against Proprietary Marketplaces
Strategically, the announcement highlights two contrasting models for exchanging AI assets. The first, a proprietary model, confines agent skills and models within a vendor’s marketplace, leading to potential vendor lock-in and associated exit costs. The second model is open, based on a neutral and documented protocol that any provider can replicate. By placing OpenSharing under the governance of the Linux Foundation, known for hosting open standards that have shaped past generations of internet and cloud infrastructure, Databricks firmly aligns OpenSharing with the latter camp.
This initiative builds upon Databricks’ previous commitments to the Linux Foundation, such as the governance of its model lifecycle management project. Handing over protocol governance to a neutral foundation shifts the control point away from the contributing vendor, creating a compelling distinction against closed marketplaces. The success of the standard will depend on the effectiveness of this open governance and the willingness of a diverse range of stakeholders to implement it independently of Databricks, expanding beyond the initial circle of Delta Sharing adopters.
A Layer of Interoperability for the Agent Ecosystem
OpenSharing aspires to serve as a common interoperability layer within the emerging ecosystem of AI agents. Its key objective is to facilitate the seamless transfer of agents, models, and data across organizational and platform boundaries without requiring tailored integrations for every connection. This aligns OpenSharing with a family of open standards aimed at streamlining asset movement between heterogeneous systems, particularly in a landscape where numerous specific integrations hinder large-scale deployment.
For a European technical leader, the appeal of an open protocol lies in its reversibility, although the nature of the actors supporting it must be considered. Most of the organizations backing Delta Sharing, including Databricks, belong to the U.S. ecosystem, which raises questions about the jurisdiction applied to the exchanged data—especially concerning the exposure to the Cloud Act for flows passing through U.S.-controlled platforms. The protocol’s open nature does not eliminate these jurisdictional concerns, which are tied to the execution infrastructure rather than the specification itself.
Another Building Block in the Data Platform Consolidation
This announcement reflects the ongoing competition among data platforms vying for the foundational role in enterprise AI. By standardizing asset exchanges, Databricks aims to enhance the attractiveness of its ecosystem while contributing to a common technical good. The strategy combines two approaches: opening the protocol to encourage wider adoption while ensuring that Databricks remains the primary beneficiary of a standard it controls in terms of origin and reference implementation.
The coming months will be critical for the actual adoption of OpenSharing by competing providers, as well as for the Linux Foundation’s ability to foster genuinely multi-party governance. The protocol emerges in a phase where the industry is actively seeking interoperability standards for agents, with numerous concurrent initiatives already structuring various levels within the agent stack. Ultimately, the success of OpenSharing will depend less on its technical prowess and more on perceived neutrality and the scope of the ecosystem that embraces it beyond its initial contributor.



